Conjecture (org)

TagLast edit: 9 Mar 2023 2:09 UTC by Andrea_Miotti

Conjecture is an alignment startup founded by Connor Leahy, Sid Black and Gabriel Alfour, which aims to scale alignment research.

The initial directions of their research agenda include:

New frames for reasoning about large language models
Scalable mechanistic interpretability
History and philosophy of alignment

We Are Conjecture, A New Alignment Research Startup

Connor Leahy8 Apr 2022 11:40 UTC

197 points

25 comments4 min readLW link

Connor Leahy on Dying with Dignity, EleutherAI and Conjecture

Michaël Trazzi22 Jul 2022 18:44 UTC

195 points

29 comments14 min readLW link

(theinsideview.ai)

Epistemological Vigilance for Alignment

adamShimi6 Jun 2022 0:27 UTC

65 points

11 comments10 min readLW link

Simulators

janus2 Sep 2022 12:45 UTC

609 points

162 comments41 min readLW link 8 reviews

(generative.ink)

Questions about Conjecure’s CoEm proposal

Akash and NicholasKees

9 Mar 2023 19:32 UTC

51 points

4 comments2 min readLW link

Conjecture internal survey: AGI timelines and probability of human extinction from advanced AI

Maris Sala22 May 2023 14:31 UTC

154 points

5 comments3 min readLW link

(www.conjecture.dev)

Re-Examining LayerNorm

Eric Winsor1 Dec 2022 22:20 UTC

125 points

12 comments5 min readLW link

Refine’s First Blog Post Day

adamShimi13 Aug 2022 10:23 UTC

55 points

3 comments1 min readLW link

Cognitive Emulation: A Naive AI Safety Proposal

Connor Leahy and Gabriel Alfour

25 Feb 2023 19:35 UTC

194 points

46 comments4 min readLW link

Searching for Search

NicholasKees and janus

28 Nov 2022 15:31 UTC

94 points

9 comments14 min readLW link 1 review

Conjecture: a retrospective after 8 months of work

Connor Leahy, Sid Black, Gabriel Alfour and Chris Scammell

23 Nov 2022 17:10 UTC

186 points

9 comments8 min readLW link

Human decision processes are not well factored

remember and Gabriel Alfour

17 Feb 2023 13:11 UTC

33 points

3 comments2 min readLW link

Empathy as a natural consequence of learnt reward models

beren4 Feb 2023 15:35 UTC

46 points

26 comments13 min readLW link

Critiques of prominent AI safety labs: Conjecture

Omega.12 Jun 2023 1:32 UTC

12 points

32 comments33 min readLW link

AMA Conjecture, A New Alignment Startup

adamShimi9 Apr 2022 9:43 UTC

47 points

42 comments1 min readLW link

Refine’s Second Blog Post Day

adamShimi20 Aug 2022 13:01 UTC

19 points

0 comments1 min readLW link

What I Learned Running Refine

adamShimi24 Nov 2022 14:49 UTC

108 points

5 comments4 min readLW link

The Singular Value Decompositions of Transformer Weight Matrices are Highly Interpretable

beren and Sid Black

28 Nov 2022 12:54 UTC

198 points

33 comments31 min readLW link

AGI in sight: our look at the game board

Andrea_Miotti and Gabriel Alfour

18 Feb 2023 22:17 UTC

225 points

135 comments6 min readLW link

(andreamiotti.substack.com)

Christiano (ARC) and GA (Conjecture) Discuss Alignment Cruxes

Andrea_Miotti, paulfchristiano, Gabriel Alfour and OliviaJ

24 Feb 2023 23:03 UTC

61 points

7 comments47 min readLW link

Japan AI Alignment Conference

Chris Scammell and Katrina Joslin

10 Mar 2023 6:56 UTC

64 points

7 comments1 min readLW link

(www.conjecture.dev)

How to Diversify Conceptual Alignment: the Model Behind Refine

adamShimi20 Jul 2022 10:44 UTC

87 points

11 comments8 min readLW link

Mosaic and Palimpsests: Two Shapes of Research

adamShimi12 Jul 2022 9:05 UTC

39 points

3 comments9 min readLW link

Refine: An Incubator for Conceptual Alignment Research Bets

adamShimi15 Apr 2022 8:57 UTC

144 points

13 comments4 min readLW link

Circumventing interpretability: How to defeat mind-readers

Lee Sharkey14 Jul 2022 16:59 UTC

114 points

15 comments33 min readLW link

Conjecture: Internal Infohazard Policy

Connor Leahy, Sid Black, Chris Scammell and Andrea_Miotti

29 Jul 2022 19:07 UTC

131 points

6 comments19 min readLW link

Understanding Conjecture: Notes from Connor Leahy interview

Akash15 Sep 2022 18:37 UTC

107 points

23 comments15 min readLW link

Methodological Therapy: An Agenda For Tackling Research Bottlenecks

adamShimi, Lucas Teixeira and remember

22 Sep 2022 18:41 UTC

54 points

6 comments9 min readLW link

Interpreting Neural Networks through the Polytope Lens

Sid Black, Lee Sharkey, Connor Leahy, beren, CRG, merizian, Eric Winsor and Dan Braun

23 Sep 2022 17:58 UTC

144 points

29 comments33 min readLW link

Mysteries of mode collapse

janus8 Nov 2022 10:37 UTC

283 points

57 comments14 min readLW link 1 review

Current themes in mechanistic interpretability research

Lee Sharkey, Sid Black and beren

16 Nov 2022 14:14 UTC

89 points

2 comments12 min readLW link

Conjecture Second Hiring Round

Connor Leahy, Sid Black, Gabriel Alfour and Chris Scammell

23 Nov 2022 17:11 UTC

92 points

0 comments1 min readLW link

The First Filter

adamShimi and Gabriel Alfour

26 Nov 2022 19:37 UTC

67 points

5 comments1 min readLW link

Biases are engines of cognition

remember and Gabriel Alfour

30 Nov 2022 16:47 UTC

46 points

7 comments1 min readLW link

Tradeoffs in complexity, abstraction, and generality

remember and Gabriel Alfour

12 Dec 2022 15:55 UTC

32 points

0 comments2 min readLW link

Psychological Disorders and Problems

adamShimi and Gabriel Alfour

12 Dec 2022 18:15 UTC

39 points

6 comments1 min readLW link

Mental acceptance and reflection

remember and Gabriel Alfour

22 Dec 2022 14:32 UTC

34 points

1 comment2 min readLW link

Basic Facts about Language Model Internals

beren and Eric Winsor

4 Jan 2023 13:01 UTC

130 points

19 comments9 min readLW link

AGI will have learnt utility functions

beren25 Jan 2023 19:42 UTC

36 points

3 comments13 min readLW link

FLI Podcast: Connor Leahy on AI Progress, Chimps, Memes, and Markets (Part 1/3)

remember and Andrea_Miotti

10 Feb 2023 13:55 UTC

39 points

0 comments43 min readLW link

Don’t accelerate problems you’re trying to solve

Andrea_Miotti and remember

15 Feb 2023 18:11 UTC

100 points

27 comments4 min readLW link

No One-Size-Fit-All Epistemic Strategy

adamShimi20 Aug 2022 12:56 UTC

24 points

2 comments2 min readLW link

Shapes of Mind and Pluralism in Alignment

adamShimi13 Aug 2022 10:01 UTC

33 points

2 comments2 min readLW link

Abstracting The Hardness of Alignment: Unbounded Atomic Optimization

adamShimi29 Jul 2022 18:59 UTC

68 points

3 comments16 min readLW link

Levels of Pluralism

adamShimi27 Jul 2022 9:35 UTC

37 points

0 comments14 min readLW link

Robustness to Scaling Down: More Important Than I Thought

adamShimi23 Jul 2022 11:40 UTC

38 points

5 comments3 min readLW link

Gradient hacking is extremely difficult

beren24 Jan 2023 15:45 UTC

162 points

22 comments5 min readLW link

Why almost every RL agent does learned optimization

Lee Sharkey12 Feb 2023 4:58 UTC

32 points

3 comments5 min readLW link

[Interim research report] Taking features out of superposition with sparse autoencoders

Lee Sharkey, Dan Braun and beren

13 Dec 2022 15:41 UTC

149 points

23 comments22 min readLW link 2 reviews

Japan AI Alignment Conference Postmortem

Chris Scammell and Katrina Joslin

20 Apr 2023 10:58 UTC

71 points

8 comments8 min readLW link

Basic facts about language models during training

beren21 Feb 2023 11:46 UTC

97 points

15 comments18 min readLW link

A response to Conjecture’s CoEm proposal

Kristian Freed24 Apr 2023 17:23 UTC

7 points

0 comments4 min readLW link

A technical note on bilinear layers for interpretability

Lee Sharkey8 May 2023 6:06 UTC

58 points

0 comments1 min readLW link

(arxiv.org)

Shah (DeepMind) and Leahy (Conjecture) Discuss Alignment Cruxes

OliviaJ, Rohin Shah, Connor Leahy and Andrea_Miotti

1 May 2023 16:47 UTC

96 points

10 comments30 min readLW link

My guess at Conjecture’s vision: triggering a narrative bifurcation

Alexandre Variengien6 Feb 2024 19:10 UTC

75 points

12 comments16 min readLW link

A couple of questions about Conjecture’s Cognitive Emulation proposal

Igor Ivanov11 Apr 2023 14:05 UTC

30 points

1 comment3 min readLW link

Barriers to Mechanistic Interpretability for AGI Safety

Connor Leahy29 Aug 2023 10:56 UTC

63 points

13 comments1 min readLW link

(www.youtube.com)

Conjecture: A standing offer for public debates on AI

Andrea_Miotti16 Jun 2023 14:33 UTC

29 points

1 comment2 min readLW link

(www.conjecture.dev)

No comments.

Con­jec­ture (org)

Conjecture (org)